building-with-jev
Pass
Audited by Gen Agent Trust Hub on Oct 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructions facilitate the processing of untrusted external data (such as user-submitted tickets, pull request bodies, and external document passages) by incorporating it into the 'state' field of Jev API judgments.
- Ingestion points: External content enters the agent's reasoning context through the 'state' field of the Jev API request, as documented in
SKILL.mdandreferences/state-and-budget.md. - Boundary markers: The skill explicitly recommends using clear delimiters (e.g.,
[Request],[Diff],[Prior Assessment]) and named constants to bound untrusted fields, as detailed inreferences/state-and-budget.md. - Capability inventory: The skill is configured with powerful tools including
Bash,Read,Edit,Write,Glob, andGrep. - Sanitization: The documentation provides specific strategies for filtering irrelevant state in code, using structured JSON schemas to prevent 'state steering', and refers to established patterns for untrusted content handling to mitigate injection risks.
- [SAFE]: The skill demonstrates high security awareness by providing instructions to keep API keys server-side, implementing a 'shadow mode' for new integrations, and establishing a 'three-tier' hierarchy that prioritizes deterministic code over model judgments.
Audit Metadata